Abstract: Geostatistical modeling technique based on seisimic data emerged in recent years, which is a new and high technique as one of the most important parts in reservoir modeling. It could realize the reservoir description and modeling and lower the uncertainty in reservoir modeling. The well log data have a high resolution in vertical, while seismic data could reflect geostructure and sand changes of the reservoir in a large scale. So high resolution reservoir modeling can be obtained by combining logging and seismic data. SGeMS'snesim module developed by Standford University was used to simulate which extends the input data to well log data, training image and sand probability. The training image was draw according to sand body isochore map, seismic inversion slice, sedimentary system of meandering river and source direction, and sand body probability was obtained by seismic data inversion. Through using the seismic data, the uncertainty of cross well region between the well was lowered and geologic recognition of reservoirs was increased using training image. The modeling result was not only reflected the shape of target structure, but also more loyal to original geologic features, especially for area without too many wells.